RTX 4090: All About Nvidia’s Lovelace GPU and Architecture

THE RTX 4090 it’s already in the oven Nvidia to be released at the end of this year. However, with several rumors floating around the internet in recent months, it is difficult to keep up with all the likely specs of the GPU and other cards to come. Ada Lovelace architecture.

With that in mind, we have prepared a guide with the main information about these products, such as technical specifications, consumption, probable performance, launch and prices.

Who was Ada Lovelace?

First of all, let’s talk about the new technology of Nvidia’s graphics cards. The company often names its architectures after great minds such as Nikolai Tesla, Blaise Pascal, Johannes Kepler and Alan Turing. This time, the chosen one was Augusta Ada Byron King, known as ada lovelace.

the mother of programmingthe mother of programmingSource: Source code/reproduction

The Countess of Lovelace is considered the world’s first programmer. In the 19th century, Ada developed the first computer program, so to speak, which was an algorithm used by a calculator called the Analytical Engine.

RTX 4090

As the first flagship of Nvidia’s new architecture, the RTX 4090 is the subject of great speculation for its high firepower. Recent rumors point out that the plaque must be 66% faster than the RTX 3090 Ticurrent most powerful card found for home computers, while the margin can rise to a whopping 82% when compared to the RTX 3090.

d NVIDIA/disclosure

The graphics card will have the AD102-300 GPU, but only the Ti model will receive the chip fully activated, according to speculation. The RTX 4090 should also have 128 stream multiprocessors and a total of 16,383 CUDA coresplus 96 MB of L2 cache memory and 384 ROPs.

The manufacturing process is done by TSMC, using 5nm lithography. Regarding frequencies, the base clock will be 2235 MHz, while the maximum clock should be around 2750 MH, and can exceed 2.8 GHz in custom models. The RTX 4090 uses 24GB GDDR6X VRAMclocked at 21 Gbps and 384-bit interface, generating a bandwidth of 1 TB.

450 and 600W versions

Regarding the consumption of this card, new signs point out that the company on the green side will launch two versions of the RTX 4090: one with 450W and the other with 600W. Both variants will have the same frequencies, but as expected, the 600W model is less likely to reach its maximum TGP and, consequently, should provide greater performance to the player. In addition, third-party video card manufacturers, such as ASUS and Gigabyte, must state on the packaging the number of watts of the cards.


The release of the RTX 4090 is really a big unknown, as several reliable industry sources point to different windows on some occasions, and Nvidia has been delaying the arrival of these cards internally. However, the month of October turns out to be the most plausible guess if we look at the company’s recent history.

Regarding the price, RTX 4090 should have a lower price than the RTX 3090 Ti, which had a suggested retail price of $1,999. However, this also means that the new generation will be more expensive than the “normal” RTX 3090, which was $1499 at launch.

RTX 4090 Ti

The big sister of the previous plate, the RTX 4090 Ti should be further away from release in the coming months, however, that hasn’t stopped famous insiders from revealing the monstrous specs of this GPU.

May information pointed to an Nvidia card that would also use the AD102 chip, with a TGP of 900W and incredible 48 GB of GDDR6X VRAM memory and clocked at 24 Gbps. Previously, it was believed that this card would be the “normal” RTX 4090, however, as time went by it became clear that this would be the Ti model.

Still in the month of May, photos of the plate’s heat sinks began to circulate on the internet. The heat sink has a sturdy structure, but what stands out is the increase in the silver baseplate, that is, the metal plate that comes into contact with the chip to carry out heat exchanges. It is interesting to note that the body of VRAM memories will also be protected.

The Chipwell Forums

But it’s not all roses, and the same insider who revealed the possible specs of this board revealed that there was still no confirmation whether this model would really see the light of day, or die with just an internal test.

RTX 4080 and RTX 4070

Finally, the second part of the high-end segment also had several leaked specifications in recent times. According to the same insiders, the RTX 4080 will have 16GB of GDDR6X VRAM, also clocked at 21Gbps, but utilizing a 256-bit interface and 672GB/s bandwidth.

The TDP should be 420W, considering the reductions in video memory specs. The chip used will be the AD103-300, which has 10,240 CUDA cores and 80 stream multiprocessors. The total ROPs is up to 224 while the L2 cache will be 64 MB.

The RTX 4070 is more modest, however, it should be able to reach 4K with ease thanks to 6144 CUDA cores and up to 56 stream multiprocessors, in addition to 160 ROPs and 46 MB of L2 cache. The 10GB memory uses the GDDR6X standard, clocked at 18Gbps and has a 160-bit interface. The TDP will be 300W.


If with the RTX 4090 it was already difficult to predict a concrete window, with these two GPUs the situation tends to be even more complicated. Apparently, Nvidia is afraid to launch both cards at the end of the year and have large stocks of the RTX 30XX generation stuck, and that’s why it announced the reduction in the value of these cards.

Thus, the most plausible scenario would be to wait for these GPUs in the month of January or February 2023. Information about the possible price of the cards was not disclosed.

Due to the appearance of various information in recent months, the RTX 40XX generation and the Ada Lovelace architecture should yield good results to players in the coming months.

Source link

About Admin

Check Also

Instagram founders announce TikTok-style text-focused social network

Instagram co-founders Kevin Systrom and Mike Krieger announced the launch of a social network called …

Leave a Reply

Your email address will not be published. Required fields are marked *